ESTRO 2023 - Abstract Book

S263

Saturday 13 May

ESTRO 2023

Conclusion The AI model successfully created a treatment plan in 67% of the cases and after small manual adjustments, 87% of the cases did fulfill all clinical goals. 50% of the manual adjustments did not decrease mean heart or lung dose. These results leave room for improvement of both training of the AI model to increase the number of cases fulfilling all clinical goals, as well as education of RTTs on clinically relevant adjustments. PD-0329 Evaluation of a CT scanner based deep learning auto-contouring solution for prostate radiotherapy M. Williams 1 , S. Berenato 1 , O. Woodley 1 , C. Möhler 2 , E. Evans 3 , A. Millin 1 , P. Wheeler 1 1 Velindre Cancer Centre, Radiotherapy Physics, Cardiff, United Kingdom; 2 Siemens Healthineers, ,, Forchheim, Germany; 3 Velindre Cancer Centre, Medical Directorate, Cardiff, United Kingdom Purpose or Objective For extreme hypo-fractionated prostate radiotherapy, geometrically and dosimetrically evaluate DirectORGANS: a novel commercial AI solution that is natively integrated into a CT scanner and utilises dedicated reconstructions optimised and standardised for auto-contouring. Materials and Methods CT scans of 20 prostate patients were sequentially selected to evaluate AI contouring for rectum, bladder and proximal femurs. 5 plan generation ‘pipelines’ were considered. 3 used AI contours with differing levels of manual editing: nominally none (AI -Std ), minor editing in specific regions e.g. target/OAR boundaries (AI- MinEd ), and fully corrected (AI -FullEd ). The remaining 2 were manual delineations from different observers (MD -Ob1 ,MD -Ob2 ). MD- Ob1 was defined as the reference contour set in all analysis. Contouring time was recorded and plans generated for each pipeline using a validated automated planning solution. The geometric and dosimetric agreement of contour sets AI- Std , AI- MinEd , AI -FullEd and MD -Ob2 were evaluated against the reference set MD- Ob1 . The non-inferiority of the AI pipelines was assessed with the testing hypothesis that ‘absolute deviations in geometry and dose metrics for AI contouring (vs MD- Ob1 ) were no greater than that from a second observer (MD- Ob2 )’. For dosimetric comparison the error in Reported Dose (RD) and Patient Dose (PD) was evaluated. RD was defined as DVH parameters that would be reported in patient records for a given pipeline. The dose distribution of each pipeline plan was evaluated on both the reference (RD- Ref ) and pipeline (RD- Pipe ) contour sets, with the difference calculated to assess the impact of contour discrepancies on RD. PD, defined as the best estimate of the actual dose the patient would receive, was extracted from the pipeline plan’s DVH using the reference contour set (PD -Pipe ). By comparing PD- Pipe with plans generated by and evaluated using MD- Ob1 (PD- Ref ), a contour set’s influence on the optimisation process and hence final dose distribution, was assessed. Results Compared to MD- Ob1 , overall delineation time for AI -Std , AI- MinEd and AI- FullEd was reduced by 24.9min (96%), 21.4min (79%) and 12.2min (45%) respectively. AI- Std contours exhibited good geometric alignment to MD- Ob1 (Figure1) with median DSC results of 0.89, 0.95, 0.96 and 0.95 for rectum, bladder, femur_R and femur_L respectively. Minor editing led to marginal improvements but both AI- Std and AI- MinEd DSC results were statistically inferior to MD- Ob2 .

All pipelines exhibited generally good dosimetric agreement with MD- Ob1 . For RD median deviations were within ±1.8cm 3 , ±1.7% and ±0.6Gy for absolute volume, relative volume and mean dose metrics respectively (Figure2). For PD, agreement was improved with respective values within ±0.4cm 3 , ±0.5% and 0.2Gy. Statistically AI- MinEd and AI- FullEd were dosimetrically non-inferior to MD- Ob2 .